This paper proposes an attack intent inference framework for defending against hypersonic glide vehicles (HGVs). Predicting the HGV behaviors poses significant challenges for defense systems due to their highly dynamic and erratic maneuvers. Complementing the limitations of the dynamics model, a unified dynamics and decision-making model of HGV is developed. First, dynamically feasible attack regions can be set by the dynamics model. Within this region, the decision-making model encodes the rational intent of attack, strategically selecting the target that maximally attains the threat value. To further address the dynamical uncertainties and potential discrepancies from the rational decision-making model, a proximity parameter is introduced in light of the Maximum Entropy principle. The attack intent of the HGV is then inferred by the Bayesian approach, whereby recursively updates the probability of the potential target to be attacked. Numerical simulations demonstrate that the proposed framework achieves superior accuracy and faster convergence in intent inference compared to existing methods, under different scenarios with varying uncertainty levels.